JP6790889B2 - Temperature prediction system and temperature prediction method - Google Patents

Description

This case relates to a temperature prediction system and a temperature prediction method.

A technique for measuring the temperature distribution in a predetermined region by acquiring the temperature of each measurement point in the predetermined region using an optical fiber or the like is disclosed (see, for example, Patent Documents 1 and 2).

International Publication WO2010-125712 Japanese Unexamined Patent Publication No. 2010-160081

In a space such as a server room where a plurality of heat sources are accommodated, a technique for estimating a future high temperature location from the heat generation amount of the heat source and the current temperature distribution is desired. For example, when predicting the future temperature of a specific measurement point using the above measurement technique, a prediction model that predicts the temperature of the specific measurement point by referring to the current temperature and calorific value of the surrounding measurement points is used. It is conceivable to use it. However, in this case, if the temperature information that does not originally need to be referred to is strongly influenced, the temperature prediction accuracy may decrease.

In one aspect, the present invention aims to provide a temperature prediction system and a temperature prediction method capable of accurately predicting future temperatures.

In one embodiment, the temperature prediction system includes a temperature sensor that measures the temperature of a plurality of measurement points in a predetermined region, an acquisition unit that acquires the calorific value of a heat generating source in the predetermined region, and measurement of the plurality of measurement points. Using a model that predicts the temperature of the measurement point of the temperature prediction target after the predetermined time by referring to the temperature and the calorific value acquired by the acquisition unit, the temperature of the measurement point of the temperature prediction target after the predetermined time is determined. A prediction unit for predicting is provided, and the parameters of the model are set with weights that refer to the measurement temperatures of the plurality of measurement points according to the positional relationship between the measurement points.

The future temperature can be predicted accurately.

It is a figure which illustrates the outline of the air conditioning management system. It is a figure which illustrates the outline of the air conditioning management system. It is a figure which illustrates the whole structure of the temperature prediction system which concerns on Example 1. FIG. It is a figure which illustrates the whole structure of the temperature prediction system which concerns on Example 1. FIG. It is a figure which illustrates the exchange of information between each part of a control device. It is a figure which illustrates the input / output of the prediction model of each measurement point. An example of a mathematical formula representing the least squares estimation is illustrated. (A) and (b) are diagrams showing the predicted temperature predicted by using the prediction model calculated by the least squares estimation and the measured temperature at the measurement point. An example of a mathematical formula representing the least squares estimation is illustrated. (A) to (c) are diagrams illustrating a range to be referred to when creating a model. It is a block diagram for demonstrating the hardware configuration of a control device. It is a figure exemplifying another example of a temperature prediction system.

Prior to the description of the embodiment, the outline of the air conditioning management system will be described as an example of the system to which the temperature prediction system is applied. 1 and 2 are diagrams illustrating an outline of the air conditioning management system 200. With the advent of the highly information-oriented society, a large amount of data is being handled by servers (computers). As a result, a plurality of servers are often installed in the same room for centralized management. For example, as illustrated in FIGS. 1 and 2, in a data center, a plurality of racks 11 are installed in a server room, and one or more servers 12 are stored in each rack 11. In the example of FIG. 1, No. 1-No. Ten racks 11 are arranged. Then, the jobs are organically distributed to the plurality of servers 12, and a large number of jobs are efficiently processed.

A large amount of heat is generated from the server 12 as the job is processed. Therefore, in order to avoid failure, malfunction, reduction of processing capacity, etc. of the server 12 due to heat, a means for cooling the server 12 is required.

For example, as illustrated in FIG. 2, the server room of the data center is an underfloor space where the equipment installation area 10a where the rack 11 is installed and the underfloor space where the power cable, the communication cable, etc. are arranged under the floor of the equipment installation area 10a. It is separated into 10b. Low temperature air is supplied from the air conditioner 13 to the underfloor space 10b. This low-temperature air is sent to the equipment installation area 10a by the air conditioning fan 20 installed in the underfloor space 10b through the ventilation port 14 provided on the floor of the equipment installation area 10a.

In the equipment installation area 10a, a plurality of racks 11 are arranged side by side in each row. For example, low-temperature air is introduced from the front surface of the rack 11 to cool the server 12, and the air whose temperature has risen is discharged from the back surface. Hereinafter, the front surface (intake side surface) of the rack 11 is referred to as an intake surface, and the back surface (exhaust side surface) of the rack 11 is referred to as an exhaust surface. The high-temperature air discharged from the exhaust surface of the rack 11 is sent to the air conditioner 13 via a pipe 15 provided on the ceiling or the like, cooled, and reused as low-temperature air.

By the way, from the viewpoint of energy saving and prevention of global warming, there is a demand for reduction of electric power consumed in a data center. Since a large amount of electric power is consumed to cool the server 12 in the data center, the air conditioning management system 200 itself is reduced in electric power, and the rack 11 is arranged in a devised manner for efficient cooling. For example, in a general data center, a large number of racks 11 are arranged in rows and arranged between racks 11 in adjacent rows so that the intake surface and the intake surface or the exhaust surface and the exhaust surface face each other. A ventilation port 14 is arranged on the floor on the side.

In this way, the cooling efficiency is improved by spatially separating the area where the low temperature air is supplied through the ventilation port 14 and the area where the high temperature air is discharged from the rack 11. .. The area on the intake surface side of the rack 11 to which low temperature air is supplied is called a cold aisle. The area on the exhaust surface side of the rack 11 from which high-temperature air is discharged is called a hot aisle.

However, high-temperature exhaust gas wraps around from the hot aisle to the cold aisle side, and a portion (hot spot) that becomes locally hot may occur, resulting in unstable equipment operation. In order to avoid this, it is conceivable to lower the set temperature of the air conditioner 13 so that hot spots do not occur even if the exhaust gas wraps around to the cold aisle side, or the exhaust gas does not wrap around to the cold aisle side. As described above, it is conceivable to set a large amount of air blown out from the air conditioner 13.

However, in either method, if the setting of the air conditioner that cools the whole is changed in order to cope with some hot spots, the area other than the hot spots will be excessively cooled. In this case, the electric power required for air conditioning increases, and energy is wasted. It is conceivable to provide an air conditioner for each rack 11, but the number of air conditioners installed will increase. Therefore, an air-conditioning management system that cools the hotspot by locally supplying low-temperature air to the hotspot by adjusting the air-conditioning direction and the amount of air-blowing of the air-conditioning fans 20 which are smaller than the rack 11. Is conceivable to build.

In such an air conditioning management system, it is desirable to predict and identify in advance where hot spots will appear. In the following examples, a temperature prediction system and a temperature prediction method capable of accurately predicting future temperatures will be described.

3 and 4 are diagrams illustrating the overall configuration of the temperature prediction system 100 according to the first embodiment. As illustrated in FIG. 3, the temperature prediction system 100 includes an air conditioning fan 20, a temperature sensor 30, a control device 40, and the like. The air conditioning fan 20 has the same function as the air conditioning fan 20 described with reference to FIGS. 1 and 2. The temperature prediction system 100 is applied to the server room described with reference to FIGS. 1 and 2 as an example.

As illustrated in FIG. 4, the temperature sensor 30 has a plurality of measurement points 31 in each rack 11. As an example, a plurality of measurement points 31 are set on the intake surface of each rack 11. For example, a thermometer or the like may be installed at each measurement point 31, but an optical fiber may be used. For example, the temperature of each measurement point 31 can be measured by laying an optical fiber so as to pass through each measurement point 31 and performing a correction calculation using a transfer function for the temperature distribution detected by the optical fiber. .. Thereby, the temperature distribution in the server room can be measured. As an example, the temperature of the measurement points 31 set at intervals of 10 cm to several tens of cm along the length direction of the optical fiber can be detected with high accuracy. In this embodiment, as an example, 723 measurement points 31 are provided.

A server 12 that generates heat is arranged in at least one of the racks 11. In the example of FIG. 4, No. One server 12 is arranged in the rack 11 of 2. No. 1, No. 3-5, No. Two servers 12 are arranged in 7 to 10. No. The server 12 is not arranged in the rack 11 of 6. Therefore, a total of 17 servers 12 are arranged. The positions of the servers 12 are D _{1 to} D ₁₇ .

The control device 40 functions as a temperature acquisition unit 41, a calorific value acquisition unit 42, a temperature prediction unit 43, a control unit 44, and the like. The temperature acquisition unit 41 acquires the measurement temperature at each measurement point 31. The calorific value acquisition unit 42 acquires the calorific value of each server 12. For example, the calorific value acquisition unit 42 acquires the calorific value of each server 12 from the measurement result of the wattmeter that measures the power used by each server 12. When the amount of heat generated by the server 12 increases, hot spots tend to appear. Therefore, the temperature prediction unit 43 refers to the measurement temperature of each measurement point 31 acquired by the temperature acquisition unit 41 and the calorific value acquired by the calorific value acquisition unit 42, and the temperature of each measurement point 31 after a predetermined time. Predict. Thereby, the temperature prediction unit 43 identifies a future hotspot. The control unit 44 controls the air-conditioning fan 20, the air-blowing direction, the air-blowing amount, and the like so that low-temperature air is locally supplied to the hot spot specified by the temperature prediction unit 43. When a plurality of air conditioning fans 20 are provided, the control unit 44 controls the air blowing direction, the air blowing amount, and the like for each air conditioning fan 20.

FIG. 5 is a diagram illustrating the exchange of information between each part. FIG. 6 is a diagram illustrating input / output of the prediction model of each measurement point 31. As illustrated in FIGS. 5 and 6, the temperature acquisition unit 41 inputs each measurement temperature y (t) of the measurement points 1 to 723 acquired from the temperature sensor 30 into the prediction model used by the temperature prediction unit 43. Further, the calorific value acquisition unit 42 inputs each calorific value d (t) of positions D _{1 to} D ₁₇ acquired from the power meter of each server 12 into the prediction model used by the temperature prediction unit 43. “T” represents the current time. “T + Δt” represents Δt hours after the current time.

As illustrated in FIG. 6, the temperature prediction unit 43 uses a prediction model to set the calorific value d (t) at positions D _{1 to} D ₁₇ and the measurement temperatures y (t) at measurement points 1 to 723. The predicted value y (t + Δt) after Δt of the temperature of each measurement point 31 is output with reference to. For example, the predicted value y (t + Δt) can be represented by A _wy (t) + B _ud (t). “U” represents a level such as an air-conditioning fan's air-blowing direction and air-blowing amount. _Ayu (t) represents the thermal equilibrium between the measurement points 31. B _ud (t) represents a temperature rise due to heat generation of the server 12. By obtaining the predicted value y (t + Δt) of each measurement point 31, future hot spots can be specified. The control unit 44 determines the control value u ^* (t) of the air conditioning fan 20 by applying the predicted temperature value of each measurement point 31 to the evaluation function. The air conditioning fan 20 blows air based on the control value given by the control unit 44. As a result, low-temperature air is locally supplied in advance to hot spots that will appear in the future. As a result, the hot spot can be cooled efficiently.

In order to cool the hotspot efficiently, an accurate temperature prediction model is desired. As a method of constructing a mathematical model that predicts the temperature distribution, the model parameters of the mathematical model are adjusted to match the actual measurement data by utilizing the past actual measurement temperature data (the difference between the actual measurement value and the estimated value by the mathematical model is small). There is a method of deriving by the least squares method. FIG. 7 illustrates a mathematical formula representing a least squares estimation.

In the above-mentioned prediction model y (t + Δt) = _Au y (t) + _Bu d (t), _Au and _Bu correspond to model parameters. A _u can be expressed by _{[a ji]. Au} is determined for each level such as the blowing direction and the blowing amount of the air conditioning fan 20. Further, since the temperature of another measurement point is referred to when predicting the temperature of a specific measurement point, a coefficient for referring to the temperature of the other measurement point is required. Therefore, "j" is a number of each measurement point 31, and is 1 to 723 in this embodiment. “I” is the number of each measurement point 31 to be referred to, and is 1 to 723 in this embodiment. _Au indicates that this [a _ji ] is determined for each level such as the air-conditioning fan 20's air-blowing direction and air-blowing amount. B _u can be expressed by _{[b jk].} “J” is the number of each measurement point 31, and is 1 to 723 in this embodiment. “K” is the number of the server position to be referred to, and is 1 to 17 in this embodiment. _Bu indicates that this [b _jk ] is determined for each level such as the air blowing direction and the air blowing amount of the air conditioning fan 20.

_{y j (Δt) ~y j (} t n) is the measurement point j, represents the temperature measured value at each time until Derutati～t _n in the case of the time zero a predetermined time. The step value from Δt to t _n is Δt. Therefore, t _n is n · Δt. By determining a _j and b _j so that the formula at the bottom of FIG. 7 is the smallest, the difference between the measured temperature and the predicted temperature becomes the smallest.

In FIG. 7, "T" means a transposed matrix. Further, "r" means an actual measurement value by the temperature sensor 30. For example, as a coefficient related to prediction, a case where a _ji is three, that is, i = 3, and b _jk is two, that is, k = 2, and j = 1 is examined. In this case, in the formula for estimating the minimum square in FIG. 7, y ₁ (Δt) and “y ₁ (0) × a ₁₁ + y ₂ (0) × a ₁₂ + y ₃ (0) × a ₁₃ + d ₁ (0) × Difference from "b ₁₁ + d ₂ (0) x b ₁₂ ", y ₁ (2Δt) and "y ₁ (Δt) × a ₁₁ + y ₂ (Δt) × a ₁₂ + y ₃ (Δt) × a ₁₃ + d ₁ (Δt) ) × b ₁₁ + d ₂ (Δt) × b ₁₂ ”,… (hereinafter, the difference up to time t _n ) so that the sum of squares is minimized, so that a ₁₁ , a ₁₂ , a ₁₃ , b ₁₁ and b ₁₂ is determined.

FIG. 8A is a diagram showing a predicted temperature predicted by using the prediction model calculated by the least squares estimation of FIG. 7 at a certain measurement point and an actually measured temperature at the measurement point. As illustrated in FIG. 8, although the predicted temperature is close to the measured temperature, it can be seen that a pulse-like error occurs in the predicted temperature as illustrated by the circles.

A _u may be represented by the matrix equation of 723 × 723, diagonal from upper left to lower right represents a factor of measuring its own temperature. When a pulse-like error occurs, the component away from the diagonal component becomes large. That is, the influence of the measurement temperature at the points where the numbers of the measurement points 31 are separated (distance is separated) becomes large. In this way, if a parameter is calculated such that the influence of the measurement temperature at a long distance, which originally has almost no influence, becomes large, it becomes a factor that the error becomes large. Therefore, in this embodiment, when determining the parameters of the prediction model, the positional relationship between the measurement points 31 in the spatial arrangement is incorporated into the constraint condition. As a result, the accuracy of determining the model parameters is improved.

For example, the importance (weight) of a _j can be adjusted by adjusting the values of e _j and H ₇₂₃ as shown in FIG. For example, as illustrated in FIG. 10 (a), the value of e _j and H ₇₂₃ to reference only the measurement points 31 within a predetermined distance in the one-dimensional direction from the position of the measuring point 31, the temperature prediction target It is preferable to adjust. Or, as illustrated in FIG. 10 (b), the value of e _j and H ₇₂₃ to reference only the measurement points 31 within a predetermined distance in the two dimensional direction from the position of the measuring point 31, the temperature prediction target It is preferable to adjust. Or, as illustrated in FIG. 10 (c), the value of e _j and H ₇₂₃ to reference only the measurement points 31 within a predetermined distance in the three-dimensional direction from the position of the measuring point 31, the temperature prediction target It is preferable to adjust. Or, it is preferable to adjust the value of e _j and H ₇₂₃ to reference only the measurement point 31 in the rack 11 to the measurement point 31 that the focused is located.

Alternatively, the reference weights of the other measurement points 31 may be changed when predicting the temperature of each measurement point 31. For example, the reference weight of the measurement point 31 at a position close to the measurement point 31 whose temperature is to be predicted may be increased, and the reference weight of the measurement point 31 at a position far from the measurement point 31 may be decreased.

FIG. 8 (b), the measurement point 31 of the number 100, the _{e j, j-2 = 0.2} , j-1 = 0.5, j = 1, j + 1 = 0.5, j + 2 = 0.2 It is a figure showing the predicted temperature estimated using the prediction model when only the diagonal component is 1 and the others are zero for H ₇₂₃ . In other words, it is the predicted temperature when the model is created so that its own weight is 1, the weights on both sides are 0.5, and the weights on both sides are close to 0.2. Under such conditions, the weight of the measurement point at a distance close to the measurement point of the temperature prediction target becomes large, and the weight of the measurement point far from the measurement point of the temperature prediction target becomes small. As illustrated in FIG. 8B, it can be seen that the predicted temperature is close to the measured temperature.

FIG. 11 is a block diagram for explaining the hardware configuration of the control device 40. As illustrated in FIG. 11, the control device 40 includes a CPU 101, a RAM 102, a storage device 103, an interface 104, and the like. Each of these devices is connected by a bus or the like. The CPU (Central Processing Unit) 101 is a central processing unit. The CPU 101 includes one or more cores. The RAM (Random Access Memory) 102 is a volatile memory that temporarily stores a program executed by the CPU 101, data processed by the CPU 101, and the like. The storage device 103 is a non-volatile storage device. As the storage device 103, for example, a ROM (Read Only Memory), a solid state drive (SSD) such as a flash memory, a hard disk driven by a hard disk drive, or the like can be used. When the CPU 101 executes the program stored in the storage device 103, each part of the control device 40 is realized in the temperature prediction system 100. The control device 40 may be hardware such as a dedicated circuit.

(Other examples)
FIG. 12 is a diagram illustrating another example of the temperature prediction system. As illustrated in FIG. 12, the temperature prediction system has a configuration in which the air conditioning fan 20 and the temperature sensor 30 are connected to the cloud 302 through a telecommunication line 301 such as the Internet. Further, the temperature prediction system may include a monitoring server 303 that is connected to the telecommunication line 301 and monitors the server room in which the temperature sensor 30 is installed.

The cloud 302 includes the CPU 101, RAM 102, storage device 103, interface 104, and the like shown in FIG. 11, and realizes the function as the control device 40. In such a temperature prediction system, for example, the measurement result measured in a foreign server room is received by the cloud 302 installed in Japan, and the future hot spot is predicted by the cloud 302. In addition, instead of the cloud 302, a server connected via an intranet or the like may be used.

In each of the above examples, the temperature sensor 30 functions as an example of a temperature sensor that measures the temperature of a plurality of measurement points in a predetermined region. The calorific value acquisition unit 42 functions as an example of an acquisition unit that acquires the heat generation amount of the heat generation source in the predetermined region. The temperature prediction unit 43 uses a model that predicts the temperature of the measurement point to be temperature-predicted after a predetermined time by referring to the measurement temperature of the plurality of measurement points and the calorific value acquired by the acquisition unit. It functions as an example of a prediction unit that predicts the temperature of the measurement point to be predicted after the predetermined time. Further, the temperature prediction unit 43 uses the measurement temperature of the plurality of measurement points in the past, the calorific value acquired by the acquisition unit, and the measurement temperature of the plurality of measurement points after a predetermined time from the past to obtain the model. It functions as an example of the creation unit to be created.

Although the examples of the present invention have been described in detail above, the present invention is not limited to the specific examples, and various modifications and modifications are made within the scope of the gist of the present invention described in the claims. It can be changed.

20 Air conditioning fan 30 Temperature sensor 40 Control device 41 Temperature acquisition unit 42 Calorific value acquisition unit 43 Temperature prediction unit 44 Control unit 100 Temperature prediction system

Claims (8)

A temperature sensor that measures the temperature of multiple measurement points in a predetermined area,
An acquisition unit that acquires the amount of heat generated by the heat generation source in the predetermined region, and
Using a model that predicts the temperature of the measurement point of the temperature prediction target after a predetermined time by referring to the measurement temperature of the plurality of measurement points and the calorific value acquired by the acquisition unit, the measurement point of the temperature prediction target A prediction unit that predicts the temperature after a predetermined time is provided.
The temperature prediction system is characterized in that the parameters of the model are set with weights that refer to the measurement temperatures of the plurality of measurement points according to the positional relationship between the measurement points. The parameter is characterized in that the weight of the measurement point at a distance close to the measurement point of the temperature prediction target is large and the weight of the measurement point at a distance far from the measurement point of the temperature prediction target is small. The temperature prediction system according to claim 1. A creation unit that creates the model from the measurement temperatures of the plurality of measurement points in the past, the calorific value acquired by the acquisition unit, and the measurement temperatures of the plurality of measurement points after a predetermined time from the past is provided.
When creating the model, the creating unit increases the reference weight of the measurement point at a distance close to the measurement point of the temperature prediction target, and increases the reference weight of the measurement point at a distance far from the measurement point of the temperature prediction target. The temperature prediction system according to claim 1 or 2, wherein the temperature is made smaller. The temperature prediction system according to claim 3, wherein the creation unit refers only to measurement points within a predetermined distance from the measurement points to be temperature-predicted when the model is created. Multiple heat sources are provided in the predetermined area,
A plurality of accommodating portions for accommodating one or more of the heat sources are provided.
Two or more of the plurality of measurement points are set for each of the plurality of accommodating portions.
The temperature prediction system according to claim 3, wherein the creating unit refers only to the measurement points of the accommodating unit in which the measurement points of the temperature prediction target are set when creating the model. The temperature prediction system according to any one of claims 1 to 5, wherein the temperature sensor is an optical fiber that passes through the plurality of measurement points. Equipped with a blower that blows air into the predetermined area
The temperature prediction system according to any one of claims 1 to 6, wherein the blower adjusts at least one of a blower amount and a blower direction according to a prediction result of the prediction unit. The temperature sensor measures the temperature of multiple measurement points in a predetermined area,
The acquisition unit acquires the calorific value of the heat generation source in the predetermined region, and
Using a model that predicts the temperature of the measurement point of the temperature prediction target after a predetermined time by referring to the measurement temperature of the plurality of measurement points and the calorific value acquired by the acquisition unit, the measurement point of the temperature prediction target The prediction unit predicts the temperature after the predetermined time,
A temperature prediction method in which the parameters of the model are set with weights that refer to the measurement temperatures of the plurality of measurement points according to the positional relationship between the measurement points.

Images